This project aims to fill the gap between our understanding of neural computations at the algorithmic level and our understanding of the same computations at the level of cellular mechanisms ? i.e., synapses, channels, and patterns of connectivity between individual neurons. Studies of simple networks are especially powerful in filling this gap. In this proposal, we will focus on two relatively simple networks, the Drosophila antennal lobe and lateral horn. The antennal lobe is the first brain region of the olfactory system, and the lateral horn is the brain region that receives the majority of antennal lobe axonal projections. The antennal lobe is a useful model for studying the cellular mechanisms of two fundamental neural computations, gain control and temporal filtering. The lateral horn is a useful model for studying the cellular mechanisms of pattern recognition ? in this case, patterns of activity across antennal lobe glomeruli. We will investigate three main questions, each relating to the relationship between cellular elements and computations within these networks. First, why are inhibitory interneurons in the antennal lobe so diverse? These interneurons respond selectively to odor concentration increases or decreases (ON or OFF cells), or particular odor pulse repetition rates (fast or slow cells), and they are also sensitive to odor concentration over different ranges. To test the hypothesis that different interneurons have distinct computational functions, we will use large-scale serial section EM in combination with in vivo electrophysiology and optogenetics. Second, how do lateral horn neurons sample the space of olfactory glomeruli? Each lateral horn neuron receives feedforward excitation from ~4 glomeruli on average (out of 50 glomeruli in total), but the total number of postsynaptic lateral horn neurons is much smaller than the number of possible glomerular combinations, raising the question of what might be special about the glomerular combinations that actually wire together in a stereotyped fashion in every lateral horn. To test the hypothesis that there are strong statistical regularities governing which glomeruli wire together, we will use 2P-mediated optogenetic stimulation of identified glomeruli, in combination with whole cell recordings from postsynaptic lateral horn neurons. Third, how do lateral horn neurons integrate their synaptic inputs? Pattern recognition can be more powerful if it involves multiple nonlinear steps, but we do not know such nonlinearities would actually be implemented at the level of cellular mechanism. To test the hypothesis that lateral horn synaptic integration involves multiple nonlinear elements, we will use both in vivo voltage imaging and in vivo whole cell recordings, together with targeted perturbations of synaptic inhibition. As a whole, these studies should substantially advance our understanding of the cellular and synaptic building-blocks underlying fundamental neural computations. Our long-term goal is to develop a fairly complete understanding of these simple networks on multiple levels of granularity. We expect our discoveries to yield testable hypotheses and conceptual approaches which will accelerate progress in understanding more complex networks.